Electroencephalography signals-based sparse networks integration using a fuzzy ensemble technique for depression detection

Today, depression is a psychological condition that affects many individuals globally and, if untreated, can negatively impact one's emotions and lifestyle quality. Machine learning (ML) techniques have lately been used to identify mental illnesses using Electroencephalography (EEG) data. These signals are difficult and time-consuming to interpret visually because they are intricate, non-static, and irregular. As a result, computer-assisted early depression detection systems are highly desirable. The study proposes a feature extraction method for each EEG signal channel by building a sparse graph from the complete complex network using a k-round minimum spanning tree. The subjects in the dataset depict the graph's nodes, and their relationship represents the edge weights, which are determined using the Euclidean distance. Then, features from the sparse graph are extracted using the Node2vec approach and fed into classifiers to get a probability score. Finally, a fuzzy ensemble strategy is exploited at the decision level for integrating probability scores to distinguish depressed subjects from healthy people. Several experiments comparing the proposed method to seven other approaches on four publicly available datasets demonstrate the importance and superiority of the proposed strategy. The K-Nearest Neighbor classifier used in the suggested method produces the highest classification accuracy across the four datasets, with scores of 0.916, 0,960, and 0.940 respectively.